dcdevaluation

dcdevaluation (Decode - Evaluation package)

dcdevaluation is a library that optimizes the model evaluation process. It was built based on the necessities of the data science team of a company named Decode, based on São Paulo - Brazil. As it is in it's early stages, for now, it only supports classification models.

Supported metrics

• KS
• ROC AUC
• F1-Score
• Precision
• Recall
• Accuracy
• "Bad rate graph"

Requirements

• scikit-learn (>0.23.2)
• Pandas (>1.0.5)
• Matplotlib (>3.3.0)
• Numpy(>1.18.5)

User installation

• Installing

pip install dcdevaluation

• Importing

from dcdevaluation import Evaluators

Methods

Init

To use the "Evaluators" class, instanciate it into a python object passing a list of probabilities (will be outputed by your model) and the true value of your data base (target feature used to train your model)

• E.g.:

train_dataset = Evaluators(predicted_y, true_y)

evaluate()

Attributes scores for all supported metrics(see above) to your select "dataset"

train_dataset.evaluate()

This method returns:

train_dataset.ks
train_dataset.auc
train_dataset.f1
train_dataset.precision
train_dataset.recall
train_dataset.accuracy

to_table()

Creates a pandas dataframe with all supported metrics

train_dataset.to_table()

This method returns:

# DataFrame with all supported metrics
train_dataset.metric_df
# Transposed DataFrame
train_dataset.t_metric_df

split_rate_graph(bins)

Creates a graph showing the good or bad rate of your model

Has the attribute bins, which allows the user to change the desired number of splits (default = 10)

train_dataset.split_rate_graph()

find_cut(min, max)

Shows precision,recall and F1 score for 20 different cutting points.

Also has the option to select a range of cutting points (default: min = 0, max = 20)

train_dataset.find_cut()

ROC_curve(dataset)

Creates as graph showing de ROC curve and it's comparison to "the coin".

train_dataset.ROC_auc()